A broad multimodal assistant for research, creation, analysis, and agentic work.
Full evaluationGlossary · Updated Sep 2, 2026
Grounding
Anchoring a model's answers in specific, verifiable sources such as your documents, a search index, or live data.
Definition
Grounding provides the model with trusted material and instructs it to answer from that material, ideally with citations. It reduces hallucination, keeps answers current, and enables permission-aware responses in enterprise settings. Techniques include retrieval-augmented generation, web search, and structured data lookups.
Why it matters when choosing a tool
Grounded answers can be checked; ungrounded ones must be trusted. For research, support, and legal work, grounding is the feature that matters most.
Where you will meet it
AI Research & Search, AI Knowledge Management, AI Customer Support
Related terms
Retrieval-augmented generation (RAG) · Citations in AI answers · Hallucination · Knowledge base
Tools where this matters
Reviewed products in the related categories
An answer engine built around current web research and visible citations.
Full evaluation
Google's multimodal assistant with deep hooks into Search, Workspace, and Android.
Full evaluation
AI writing, search, and task assistance inside a connected workspace and knowledge base.
Full evaluation
Google's source-grounded research notebook that answers only from the documents you add and makes audio overviews.
Full evaluation
Enterprise search and an AI assistant that answer questions across a company's apps with permissions intact.
Full evaluation